IB IA 内部评估 研究设计 数据处理 评估
IA 是 IB 文凭课程里少数由本校老师打分、又能在考前几个月反复修改的部分。理科的 IA 占最终成绩 20%，上限 3000 词，总分 24 分。它考的不是实验做得多好看，而是报告能不能让阅卷员看懂你的判断。这篇按 2025 年首考的理科指南拆开讲，也带上历史、地理、经济和数学的对照口径。
The internal assessment is one of the few parts of the IB Diploma that your own teacher marks and that you can still revise months before the exams. In the sciences it is worth 20% of the final grade, capped at 3,000 words and marked out of 24. What is assessed is not how impressive the experiment looked but whether the report makes your reasoning visible. This guide follows the science criteria that were first examined in 2025 and adds the corresponding rules for history, geography, economics and mathematics.
一、各科 IA 的分值与词数上限
同一套 IB 体系里，IA 的形态在各科差别很大。生物、化学、物理的 IA 是一份科学探究报告，2025 年首考的新指南把上限压在 3000 词，占总成绩 20%，满分 24 分。数学 AA 与 AI 不看词数，要求的是 12 到 20 页的数学探究，含图表不含书目，满分 20 分，同样占 20%。历史是 2200 词上限，地理 2500 词，经济是三篇各 800 词的评论，心理学 1800 词。IBO 官网的科目说明会写清各科权重，SL 与 HL 常常不同。很多学生以为 IA 只是小作业，其实它和你某一张 paper 的分量相当。
The internal assessment looks different in every subject. In biology, chemistry and physics it is a single scientific investigation capped at 3,000 words, worth 20% of the final grade and 24 marks under the guide first examined in 2025. Mathematics AA and AI work by page count instead, with a 12 to 20 page exploration worth 20 marks and the same 20% weighting. History caps at 2,200 words, geography at 2,500, and economics uses three commentaries of 800 words each; weightings differ between SL and HL, so the subject brief on the IBO website is the version to trust.
二、理科 24 分怎么分：四项标准各 6 分
2025 年首考的理科指南把 IA 拆成四项标准，每项 6 分，合计 24 分，权重各 25%。研究设计看的是研究问题有没有放进行之有效的背景里，方法写得能不能让别人重复。数据分析看原始数据有没有被恰当处理，不确定度有没有被认真对待。结论要求回答研究问题，并和公认的科学背景做比较。评估要求指出具体的局限，再给出对应的改进。四项分开打分，所以一篇方法漂亮但数据粗糙的报告拿不到高分。
The science guide first examined in 2025 divides the investigation into four criteria of six marks each, which means every criterion carries a quarter of the total. Research design rewards a research question placed in a specific context and a method detailed enough to be reproduced. Data analysis looks at whether raw data is processed accurately and whether uncertainty is treated seriously; the conclusion has to answer the question and compare the result with accepted science, while evaluation needs specific limitations with matching improvements. Because the marks sit apart, a beautifully designed method cannot rescue careless data.
三、研究问题的背景要素：两个变量加一套理论
指南对有背景的研究问题有明确解释。它要提到自变量与因变量，或者两个相关的变量，再用一两句说明这个系统是什么。背景理论必须与研究问题直接相关，不能把课本上关于这个主题的一大段都抄上去。很多学生以为写出温度对酶活性的影响就够了，其实这只给了变量，没给系统。更稳的写法是把它落到具体对象与范围上，例如某一种酶在某个温度区间内反应速率怎么变。范围定得越清楚，后面方法里的取值间隔与重复次数才有依据。
The guide is explicit about what a research question with context contains. It has to name the independent and dependent variables, or two correlated variables, and describe the system in a sentence or two; background theory must bear directly on the question rather than summarising a whole chapter. Many students think naming the variables is enough, but that leaves the system undefined. A tighter version anchors the question in a specific organism, material or range, which is what later justifies your intervals and repeats.
四、方法论细节：范围、间隔、重复与控制变量
研究设计这一项里，方法本身占了一半的分量。要写清每种变量怎么测、数据来自哪里、取值的范围与间隔、每个点重复几次、仪器的精度。控制变量不能只列名字，还要写用什么手段控制。安全、伦理与环境问题要单独交代，涉及活体生物或人体数据时这一步不能省。常见的错误是把步骤写成流水账，却漏掉为什么这样选。指南强调的是判断，例如为什么取 5 个温度点而不是 3 个，为什么每个点做 3 次而不是 1 次。写清理由的句子，往往就是阅卷员给分的位置。
Half of research design is the methodology itself. The report should state how each variable is measured, where the data comes from, the range and interval of the independent variable, the number of repeats and the precision of the instruments. Control variables need a method of control, not just a list of names; safety, ethical and environmental issues belong in their own sentence, especially with living organisms or human data. A common mistake is writing the procedure as a recipe while leaving out the reasoning behind each choice. The upper bands reward explained decisions, such as why five temperatures rather than three, so put the reason on the page.
五、数据处理与不确定度
数据分析是最容易被低估的一项。它不只看你会不会算平均值和画图，还看你怎么表达不确定性。表格要有单位、有效数字要一致，图的坐标轴要标清变量与单位，拟合线要说明用了什么方法。不确定度可以来自仪器精度，也可以来自重复之间的离散程度，两者最好分开说明。很多学生把误差和不确定度混在一起写，整段就落在低档。指南的 5 到 6 分档要求处理恰当且准确，这意味着每个处理步骤都要有理由，而不是把软件输出直接贴上去。
Data analysis is the criterion students underestimate most; it is not only about averages and graphs but about how uncertainty is expressed. Tables need units and consistent significant figures, axes need labelled variables, and any line of best fit needs a stated method; uncertainty can come from instrument resolution or from the spread between repeats, and the two are worth separating. Many students write about error and uncertainty as if they were the same thing, which pushes the whole section into the lower bands. The top band asks for processing that is appropriate and accurate, so justify each step instead of pasting spreadsheet output.
六、结论要能和公认的科学背景对上
结论不是把结果重复一遍。它要用处理后的数据，连同不确定度，回答研究问题，再把这个回答放进已有的科学知识里比较：你的数值和文献值差多少，趋势和理论预测是否一致。指南要求引用的资料足够详细，能让读者找到原始出处。常见的错误是只写与理论相符，却没写清是哪条理论、哪个数值。结果和理论不一致时也不必藏起来，说明可能的原因反而更贴合阅卷口径。数据撑不住的结论，写得再漂亮也拿不到 5 到 6 分。
A conclusion is not a restatement of the results. It answers the research question with the processed data and its uncertainty, then places that answer against established science: how far your value sits from published values, and whether the trend matches prediction. Citations must be detailed enough for a reader to trace the source; a frequent mistake is claiming agreement with theory without naming the theory or the value. Disagreement is worth reporting honestly with a possible cause, because a claim the data cannot carry will not reach the top band.
七、评估：具体局限、对应改进、影响大小
评估这一项的分档差别最明显。1 到 2 分只要求写出通用的弱点，3 到 4 分要求具体，5 到 6 分要求解释每个局限的相对影响。通用的弱点指人为误差、设备不够精确这类放到任何实验都成立的说法。具体的局限要指向你自己的方法，例如某个温度点保温时间不足，导致读数偏低。改进要和局限一一对应，改完还要说清能改善多少。阅卷时看的是这一项里有没有真正的自我判断，而不是有没有写满一页。
Evaluation separates candidates more sharply than any other criterion; the lower band accepts generic weaknesses, the middle band wants specific ones, and the top band explains their relative impact; generic means phrases such as human error or imprecise equipment that would fit any experiment. Specific limitations point at your own method, for example insufficient equilibration time at one temperature lowering the reading; each improvement should answer one limitation and state how much it would help. What examiners look for here is real self-judgement, not a full page of text.
八、人文与社科的词数差异
人文学科的 IA 不是实验报告，口径也不同。历史要求 2200 词以内，超出的部分阅卷员不读，参考文献与标题不计入。地理是 2500 词的田野调查报告，经济是三篇各 800 词的评论，心理学 1800 词。这些上限是硬线，不是建议。很多学生以为多写能多拿分，实际上超字往往意味着筛选能力不足。IBO 官网的科目说明会写清哪些部分计入词数，动笔前先看那一页，比写完再删省时间。
In the humanities the internal assessment is not a lab report and the limits differ; history allows 2,200 words, with references and headings excluded and anything beyond the limit left unread. Geography sets 2,500 words for the fieldwork report, economics three commentaries of 800 words each, and psychology 1,800 words. These are hard limits rather than advice. Many students assume more words means more marks, but overflow usually signals weak selection; the subject brief on the IBO website lists what counts, and reading it before drafting saves a rewrite.
九、数学探究：按页数而不是词数
数学 AA 与 AI 的 IA 叫数学探究，评 20 分，占 20%。它不看词数，要求 12 到 20 页双倍行距，含图表不含书目，封面页写明页数。指南建议投入 10 到 15 小时，这个数字和实际投入常常差很远。评分看的是数学的深度与个人投入，而不是篇幅长短。常见的错误是选一个太大的主题，例如把整章微积分讲一遍，结果每个点都很浅。更有效的做法是选一个具体问题，把同一套数学用到底，并在过程中说明为什么这样建模。
The mathematics IA is a mathematical exploration marked out of 20 and worth 20% of the grade; it is judged by pages rather than words: 12 to 20 pages double spaced, figures included and bibliography excluded, with the page count on the cover. The guide suggests 10 to 15 hours of work, a figure that often differs sharply from what students actually spend. Marks come from mathematical depth and personal engagement, not length. A common mistake is choosing a topic so broad that a whole chapter of calculus is summarised at surface level; a narrower question carried through consistently, with the modelling choices explained, reads far better.
十、校内打分与校外调整：样本怎么动你的分
IA 先由本校老师按标准打分，再由 IBO 抽取样本做调整。这个机制叫 Dynamic Sampling，样本要覆盖学校的分数段，满分学生通常不被抽中，目的是给高分留上调空间。阅卷员比较样本后，如果发现学校整体偏高或偏低，调整会施加到该科目这一组件的全部考生。换句话说，你同学的分数可能一起被拉高或压低。IBO 的评估指南写明，调整后学校的分数可能上升、下降或不变。理解这一点，就知道老师为什么强调按标准写而不是按感觉写。
Your teacher marks the IA first, then the IBO moderates a sample of the work. The process is called dynamic sampling, and the sample is chosen to represent the school's mark range while usually avoiding full-mark candidates, so strong work still has room to move up. If the moderator finds the marking generous or harsh overall, one adjustment is applied to every candidate for that component. That means your mark can move together with your classmates; the IBO assessment guide states that a school's marks may be raised, lowered or left unchanged, which is why teachers insist on writing to the criteria.
十一、学术诚信与引用格式
IA 和拓展论文走同一套诚信规则。所有用到的资料都要按统一格式引用，别人采集的数据要标明来源。网上流传的范文不能抄，学校会做原创性检查，阅卷员也见过大量重复的选题。找人代写属于作弊，最重的处理可以是不发文凭。AI 工具能不能用，各校政策不同，动笔前先问清。参考文献的格式要全篇统一，别在同一份报告里混用两种。
The IA follows the same integrity rules as the extended essay; every source must be cited in one consistent style, and any data you did not collect yourself needs a visible origin. Sample reports circulating online cannot be copied: schools run originality checks and moderators have seen the popular topics many times. Commissioned work counts as malpractice, and the penalty can extend to withholding the diploma; rules on AI tools vary between schools, so ask before drafting. Keep one citation style across the whole report rather than mixing two.
十二、十小时怎么排：草稿、反馈与提交
理科 IA 在课程里分配 10 小时教学时间，包括讲要求、进实验室收数据、老师反馈与真实性核查。这 10 小时通常不够，所以真正的写作要放到课外。指南允许老师对第一稿给意见，但不允许逐段修改，边界各校会写明。一个可行的节奏是先定研究问题与方法，再收数据，然后只写数据处理与结论，最后补研究设计与评估。先写中间两节的好处是，数据的实际形态会反过来修正你的问题。提交前把词数、图表编号、引用格式对着清单过一遍。
The sciences allocate 10 hours of class time to the IA, covering the requirements, data collection, teacher feedback and authenticity checks; that is rarely enough, so the writing happens outside class. Teachers may comment on a first draft but cannot rewrite it section by section, and each school sets its own boundary. A workable order is to fix the question and method, collect data, draft analysis and conclusion, then return to design and evaluation. Starting in the middle lets the data reshape the question; before submission, run a checklist over the word count, figure numbering and citation style.
牛津高阶英汉双解词典 第10版 商务印书馆¥168📱 去京东购买京东精选
需要一对一辅导？Need One-on-One Tutoring?
IB / A-Level / GCSE 全科辅导，IA 选题与方法论可以逐段批注
16621398022 同微信
Visit aleveler.com
或截图后用「扫一扫 → 相册」选择本图